if (!CreateInput(inputs[i], cond_graph, StrCat("cond_input", i).c_str(),
                     &cond_inputs[i], status)) {
      break;
    }
    if (!CreateInput(inputs[i], body_graph, StrCat("body_input", i).c_str(),
                     &body_inputs[i], status)) {
      break;
    }
  }

  TF_WhileParams params = {ninputs,    cond_graph,  cond_inputs,  cond_output,

          downloaded_block = true;
          block->state = FetchState::FINISHED;
        } else {
          block->state = FetchState::ERROR;
        }
        block->cond_var.SignalAll();
        return;
      case FetchState::FETCHING:
        block->cond_var.WaitWithTimeout(&block->mu, absl::Minutes(1));
        if (block->state == FetchState::FINISHED) {
          return TF_SetStatus(status, TF_OK, "");
        }

  // This is the size of cond_inputs, body_inputs, and body_outputs.
  const int ninputs;

  // The while condition graph. The inputs are the current values of the loop
  // variables. The output should be a scalar boolean.
  TF_Graph* const cond_graph;
  const TF_Output* const cond_inputs;
  TF_Output cond_output;

        op_(nullptr),
        thread_(tensorflow::Env::Default()->StartThread(
            tensorflow::ThreadOptions(), "parallel_device_execute",
            std::bind(&DeviceThread::Run, this))) {}
  ~DeviceThread();

  // Requests that the worker thread execute the specified operation. Blocks
  // until the previously pending operation (a StartExecute without a Join) has

  std::vector<int64_t> targets;
  std::vector<Gradient*> used_in_grads;
  // We may end up with slightly fewer elements than we reserve, but grad.size()
  // should be a reasonably tight upper bound.
  targets.reserve(grad.size());
  used_in_grads.reserve(grad.size());
  std::unordered_map<int64_t, TapeTensor> sources_that_are_targets;
  for (int grad_index = 0, end = grad.size(); grad_index < end; ++grad_index) {

    uint64_t timestamp;
    /// Mutex to guard state variable
    absl::Mutex mu;
    /// The state of the block.
    FetchState state ABSL_GUARDED_BY(mu) = FetchState::CREATED;
    /// Wait on cond_var if state is FETCHING.
    absl::CondVar cond_var;
  };

  /// \brief The block map type for the file block cache.
  ///
  /// The block map is an ordered map from Key to Block.
  typedef std::map<Key, std::shared_ptr<Block>> BlockMap;

      represent, as a whole, an original work of authorship. For the purposes
      of this License, Derivative Works shall not include works that remain
      separable from, or merely link (or bind by name) to the interfaces of,
      the Work and Derivative Works thereof.

      "Contribution" shall mean any work of authorship, including
      the original version of the Work and any modifications or additions

    // Create loop: while (input1 < input2) input1 += input2 + 1
    TF_Operation* less_than = LessThan(
        params->cond_inputs[0], params->cond_inputs[1], params->cond_graph, s_);
    ASSERT_EQ(TF_OK, TF_GetCode(s_)) << TF_Message(s_);
    params->cond_output = {less_than, 0};

    TF_Operation* add1 = Add(params->body_inputs[0], params->body_inputs[1],

*   `tf.cond` emits a StatelessIf op if the branch functions are stateless and
    do not touch any resources.
*   `tf.cond`, `tf.while` and `if` and `while` in AutoGraph now accept a
    nonscalar predicate if has a single element. This does not affect non-V2
    control flow.
*   `tf.while_loop` emits a StatelessWhile op if the cond and body functions are